-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_local_ue.sh
executable file
·52 lines (45 loc) · 2.93 KB
/
train_local_ue.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
#!/bin/bash
# Ensemble
# Baseline without UE directly uses the model with seed=0
# Step 1: Train one-stage models with different seeds
for i in {0..4}
do
sbatch --export=ALL,SEED="$i",DC=512 scripts/local_largest_1536/train_local_sparse_extended_long_ihn.sbatch
sbatch --export=ALL,SEED="$i",DC=256 scripts/local_largest_1536/train_local_sparse_extended_long_ihn.sbatch
sbatch --export=ALL,SEED="$i",DC=128 scripts/local_largest_1536/train_local_sparse_extended_long_ihn.sbatch
echo "$i"
done
# Step 2: Train two-stage models with different seeds
for i in {0..4}
do
# Change MODEL_FOLDER to the folder of the model with seed=0
sbatch --export=ALL,SEED="$i",DC=512,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_sthn.sbatch
sbatch --export=ALL,SEED="$i",DC=256,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_sthn.sbatch
sbatch --export=ALL,SEED="$i",DC=128,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_sthn.sbatch
echo "$i"
done
# Step 3: Train uncertainty estimation models
# TTA
# One-stage models
# Change MODEL_FOLDER to the folder of the base model
sbatch --export=ALL,DC=512,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_ihn_c_ue1r32.sbatch
sbatch --export=ALL,DC=256,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_ihn_c_ue1r32.sbatch
sbatch --export=ALL,DC=128,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_ihn_c_ue1r32.sbatch
# Two-stage models
# Change MODEL_FOLDER to the folder of the base model
sbatch --export=ALL,DC=512,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_c_ue1r32.sbatch
sbatch --export=ALL,DC=256,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_c_ue1r32.sbatch
sbatch --export=ALL,DC=128,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_c_ue1r32.sbatch
# DM
# One-stage models
# Change MODEL_FOLDER to the folder of the base model
sbatch --export=ALL,DC=512 scripts/local_largest_1536/train_local_sparse_extended_long_ihn_d.sbatch
sbatch --export=ALL,DC=256 scripts/local_largest_1536/train_local_sparse_extended_long_ihn_d.sbatch
sbatch --export=ALL,DC=128 scripts/local_largest_1536/train_local_sparse_extended_long_ihn_d.sbatch
# Two-stage models
# Change MODEL_FOLDER to the folder of the base model
sbatch --export=ALL,DC=512,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_sthn_d.sbatch
sbatch --export=ALL,DC=256,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_sthn_d.sbatch
sbatch --export=ALL,DC=128,MODEL=$MODEL_FOLDER scripts/local_largest_1536/train_local_sparse_extended_long_load_f_aug64_sthn_d.sbatch
# DE
# Use the models trained in step 1 and 2. Put trained models in local_pipeline/ensembles